An effective method to analyze variations of high-dimensional patterns over medical streams

In medical field, patterns over time-varied data streams usually imply high domain value. The variations of patterns can often be very complex and hard to evaluate. Traditional methods usually take each pattern as a whole to analyze data stream variations or only focus on one type of variation; however, few works have achieved a widely applicable resolution. This paper considers the feature of sub parts for data stream patterns and studies their variations and relationships from the perspective of multiple dimensions, to explore a comprehensive understanding for the variation history and effectively support different types of queries to help analyze the variations. This paper first decomposes patterns into different dimensions and then evaluates the variations of each dimension. After that, a data cube called VS-Cube is used to find out the variations of a single dimension as well as the relationships between different dimensions within a certain pattern. At last, a case study on disease MI over medical stream is given to demonstrate the effectiveness and efficiency of our proposed methods.